the data is available here: /work/DAT_002__TCGA_Ovarian/01_2011/vita_work/ComethExpr

Hypothesis: if expression of a gene is correlated with the first principal component of the comethylation network it is possible that this gene might be involved in epigenetic regulation of gene expression. How to test? Find correlations of comethylation PC1s with gene expression. Do GO analysis with the list of the obtained genes. Look for enrichment in epigenes.

Correlations:

mb PC1 vs Affy gene expression:

P value	Raw P values	mean	FDR, %
0.01	180048	8097.11	4.5
0.05	273649	39977.1	14.61
0.001	108442	831.3	0.76
0.005	153143	4075.3	2.6

mbc PC1 vs Affy gene expression:

P value	Raw P values	mean	FDR, %
0.001	70403	981.1	1.4
0.005	113812	4841.9	4.2
0.01	142743	9647.98	6.7
0.05	252151	47890.13	18.9

All following analyses were done with mbc. Looked at different rho thresholds to see how many correlated pairs I get:

rho	0.4	0.5	0.6	0.7
Pairs	2642	869	172	3

The top three genes correlated with modules are: CARM1 (coactivator-associated arginine methyltransferase 1; module: orchid), CDKN2A (cyclin-dependent kinase inhibitor 2A; module: khaki), EBAG9 (estrogen receptor binding site associated, antigen, 9; module: forestgreen). I searched Pubmed and couldn't find any connections between CARM1 and CDKN2A and couldn't find any connection with ovarian cancer and/or DNA methylation. Also I manually look for genes involved in epigenetic regulation in the list of genes correlation with the PCs and found quite a few when the rho cutoff is between 0.4 and 0.5. Some of the notable examples are: HDAC1, HDAC2, MBD1, MBD2, MYST1, JMJD6, DNMT1, DNMT3, UTX, TRIM28 (KAP1), BRCA1 (not an epigene but found to by hypermethylated in ovarian and breast tumors), CTBP1 and CTBP2 (found in polycomb complexes), CXX1 (binds to CpGs), E2F3/E2F4 (binding sites involves methylation), EHMT2 (histone methyltransferase G9a, known to be in the same complex with DNMT1).

Next question: how many genes correlate with a single module? Answer: many genes are correlated with a single module. Question: is a gene correlated with multiple modules? Answer: every gene is correlated with only one module (except POLR2J which is correlated with 2 modules.)

Calculation of enrichment for epigenetics related genes in the list of genes correlated with coexpression modules

Xia has provided me with a list of genes related to epigenetic regulation of transcription. These were mouse genes and I needed to find human orthologs before I could calculate the enrichment. Solution: use Bioconductor annotation packages Inparanoid. Total number of mouse genes supplied:706, total number of mouse genes mapped to human genes is 566 + 35. Sweave file that as the code for these analyses and descriptions can be found in trunk/ComethExpr/Sweave.

Finally using the list of 601 human orthologs of mouse epigenes I calculated enrichment of epigenetics related factor in the list of genes correlated with modules (PC1) of the comethylation network (mbc). Before I could do any calculation I needed to map Affy Gene Symbols to ENTREZ Gene IDs using Bioconductor "org.Hs.eg.db" package, mapped 11244 IDs out of 12042. Did the same mapping for every set of Affy genes correlated to the modules at different R values. Out of 601 epigenes I was able to find 446 on the Affy platform (out of 11244 genes).

Rho	# of Gene Symbols	Mapped to ENTREZ, unique	Epigenes	P value
0.4	2642	2327	54	0.99
0.5	869	798	49	0.00079
0.6	172	159	7	0.29

Here is the code for the hypergeometric test in R (for more details see EpigenesEnrichment.pdf in /trunk/ComethExpr/Sweave).

phyper(54,446,11250-446,2327,lower.tail=FALSE)
phyper(49,446,11250-446,798,lower.tail=FALSE)
phyper(7,446,11250-446,159,lower.tail=FALSE)

I also compiled a list of epigenetic "writers" and "erasers", excluded protein that don't have any enzymatic activity (even if they are part of the complexes) and also chromatin remodelers and performed similar analyses. My set had 80 unique epigene. Total number of those on the platform is 53.

Rho	# of Gene Symbols	Mapped to ENTREZ, unique	Epigenes	P value
0.4	2642	2327	21	0.0005
0.5	869	798	4	0.32
0.6	172	159	2	0.038

Conclusions: these analyses may allow us to claim that we have some enrichment for epigenes depending on what threshold (correlation rho) we use. In any case, the next step will be to look at the list of epigenes at some threshold and look for SNPs/CNV in the coding regions of those genes.

Browser not supported